I am glad that Natalia Cecire has published a commentary — The passion of Nate Silver (sort of) — on the recent attacks and counter-attacks on Nate Silver’s predictions for the upcoming presidential debate on his New York Times hosted blog, FiveThirtyEight. Silver was described by MSNBC’s Joe Scarborough as an “ideologue” and a “joke” who is biased in favor of Obama; and Silver’s defenders have accused his attackers as, among other things, “innumerate” and “braying idiot detractors.” On the one hand, this seems like just another pseudo-controversy ginned up in the current toxic political climate, but Cecire brings her feminist and literary critical insights.
In particular, she describes the Silver controversy as “puerile,” which she uses as a technical term to mean (white American) boyish play; specifically, in this case, with a kind of virtuosity with playing games with numbers. I remember myself playing Monopoly with Alan Kulevich when we were young teens, arguing over, bending, breaking, and rewriting the rules, trash talking one another, and I think I am not far from understanding what she means.
Silver himself has been largely silent on the controversy; it has been his attackers and his defenders who have engaged in the puerile discussion. But, as Cecire points out, Silver literally tried to turn the controversy into a game, offering to bet Scarborough $1,000 over the election, and was reprimanded by the New York Times’s public editor, Margaret Sullivan, for doing so. The editor said that Silver said he was being “half playful and half serious,” which Cecire says is “the essence of the appeal of FiveThirtyEight.”
I have just finished reading Silver’s recently published book, The Signal and the Noise, in which Silver describes, in its revealing subtitle, “Why So Many Predictions Fail — but Some Don’t.” Silver has been a successful predictor in three fields: professional baseball, online poker, and the winners of political races, especially the 2008 presidential election. The first two items are certainly games with numbers.
But Silver’s main goal in writing the book is to help a general educated audience to begin to think like a Bayesian. In the end, it’s a fairly modest attempt; in fact, the book is full of modest ambition, in which he describes the science of prediction in field after field — including stock-picking and finance, earthquakes, local weather beyond a seven-day horizon, terrorist attacks, and epidemiology — as failing to provide major insights. Other fields, such as short-term forecasts, where hurricanes will land, and Silver’s particular success stories, have been much more successful.
In one chapter, Silver describes “living in Bayesland,” where people are required to act on their Bayesian beliefs in a particular way: if they believe that something will happen with some degree of certainty, they have to make a bet with the same expected value; essentially, they need to put their money with their mouth is; to put up or shut up. I think this is what is behind Silver’s proffered bet to Scarborough; Scarborough believes the odds are about 1-1 for a Romney victory, and Silver 4-1 against; Silver would expect to make a lot of money. I think this is the serious part; the playful part is that it’s strictly for the game; the Red Cross would get the money in either case. On the one hand, this appears to be simply the gamification of prediction, but Silver’s point is different: he wants predictions to have real consequences for the people who make them. In the reputation economy of the Red Cross bet, Silver engaged in fairly skillful arbitrage — besides believing that he’d win this bet four out of five times, he won by even making the offer, and by shorting himself by having the donation go to New York relief. His reacted to the public editor’s admonition by tweeting, “I think Margaret Sullivan is a terrific Public Editor,” and then by donating $2,538 to the Red Cross (and inviting others to do the same). These, I would say, are signs of maturity. It was foolish to portray Silver as Galileo, but he is displaying a kind of grace under pressure we admire.
The factors which lead to successful Bayesian thinking include being able to accurately and quantitatively measure the underlying phenomenon at hand reliably (for example, a batter’s on-base percentage, polling demographics, or earthquake tremors). It also depends on the quality and scale of measurement; it is much more difficult to predict accurately when the scale is exponential than when it is linear; making even a small error on exponential problems (such as earthquakes) quickly leads to wildly inaccurate estimates. Unfortunately, many phenomena of interest have just the quality. We may be able to estimate that there is a 90% chance of a magnitude 8 earthquake at some location over some long time-frame, and, interestingly, we may be able to do the same for large-scale terrorist attack. But, in both cases, it is difficult to measure accurately and precisely; worse, causal theories are lacking which indicate what to measure in the first place.
Both earthquake and terrorist attack predictions are serious things. Silver relates how he was invited to a think-tank discussion on terrorism at a time inopportune for his political prediction work (right before the 2008 election), and, how inadequate he felt to the task once he arrived. I wonder if it felt more like being called up to the major leagues, or more like being called to put away childish things.
Silver’s book provides a sober assessment of both the need and the limits of prediction. Against type, he even approvingly quotes Donald Rumsfeld’s categories of “known unknowns” and “unknown unknowns,” which were widely ridiculed by liberals, but which reflect a somber reality. I suppose I wanted a happier ending to the book, a kind of Dummy’s Guide to Bayesian Reasoning that could be applied to every situation. But even though Silver strives to be a Bayesian, he is aware that limits on access to useful data, the flood of noisy data, our ability to measure, and the scale of the measurements all conspire to limit how well we can predict.
Cecire’s essay serves as a reminder, too, of the practical limitations of Bayesian reasoning. Understanding racial and gender demographics as applied to electoral politics is not the same as anti-racism or anti-sexism. Indeed, it can easily lead to increased power inequalities; the direct marketing experts and manipulators of conservative evangelicals are also expert Bayesians. As I write this post, there is an increasingly strong reaction against Sullivan’s admonition (for example, “A bet is a tax on bullshit”). Cecire prompts me to consider that bets, considered as taxes, are essentially regressive, and reminds me that other power relationships are involved; even winning at reputation in the large is a game that people who already have other kinds of power have leisure to play.
These have been a lot of fairly disparate thoughts. But let me make some predictions: if Obama wins, Silver’s reputation will increase by an irrationally large amount, if Romney wins, Silver’s reputation will decrease by an irrationally large amount. Indeed, Silver has said as much. But in the long term, his reputation for political predictions will continue to increase if he continues to apply his Bayesian and fairly open methodologies to the task.
Correction: Corrected “Nick” to “Nate”. What a weird typo — perhaps from Saint Nick or Nick Saint or Quicksilver or even Nick Adams.